**M**ultiscale **G**eographically **W**eighted **R**egression (MGWR)
=======================================
[![Build Status](https://travis-ci.org/pysal/mgwr.svg?branch=master)](https://travis-ci.org/pysal/mgwr)
[![Documentation Status](https://readthedocs.org/projects/mgwr/badge/?version=latest)](https://mgwr.readthedocs.io/en/latest/?badge=latest)
[![PyPI version](https://badge.fury.io/py/mgwr.svg)](https://badge.fury.io/py/mgwr)
This module provides functionality to calibrate multiscale (M)GWR as well as traditional GWR. It is
built upon the sparse generalized linear modeling (spglm) module.
Features
--------
- GWR model calibration via iteratively weighted least squares for Gaussian,
Poisson, and binomial probability models.
- GWR bandwidth selection via golden section search or equal interval search
- GWR-specific model diagnostics, including a multiple hypothesis test
correction and local collinearity
- Monte Carlo test for spatial variability of parameter estimate surfaces
- GWR-based spatial prediction
- MGWR model calibration via GAM iterative backfitting for Gaussian model
- MGWR covariate-specific inference, including a multiple hypothesis test
correction and local collinearity
评论1
最新资源